CAPÍTULO I.- ENERGÍA EN EDIFICIOS Y SU NORMATIVA
I.2. MARCO NORMATIVO Y SU EVOLUCIÓN HISTÓRICA
I.2.4. Real Decreto 47/2007 sobre Certificación Energética
FASTRAK-APT is a project scheduling expert system developed by Hyundai Engineering and Construction. It is used for generation, verifying, and modifying construction project PERT-CPM networks. It uses Case Based Reasoning and constraint-based reasoning to assist human project planners.
See Also: Case Based Reasoning.
Fault Tree
A fault tree is an event tree that is used to represent the possible faults in a process. It can be used as a simple analysis tool or as a control or diagnosis structure for an automated system. For example, a diagnosis system could use fault tree and associated probabilities to propose checks or repairs for a computer system or a medical condition.
FCM
See: Fuzzy Cognitive Map.
FDS
FDS is a program designed to solve certain mathematical programs using Means-Ends analysis.
See Also: Means-Ends analysis.
Feature (Attribute)
See: attribute.
Feature Analysis
An image analysis technique that decomposes the image into easily recognized parts (horizontal, vertical, or diagonal lines, curves, etc.).
Page 106
The image can then be classified or recognized by comparing the feature list with those of various standards.
Feature Extraction
Feature Extraction is used in speech recognition and image processing to refer to the reduction of the input signal into a collection of broader features that can be used for further analysis. It is used more generally as term for the process of variable reduction.
See Also: data reduction.
Feature Points
In image analysis, the feature points are a list of identifiable places observed in an image.
Feature Vector
A feature vector is one method used to represent a textual or visual object in a form suitable for numeric processing and machine learning. As an example, a block of text (e.g., an article in a newspaper) could be collapsed into a (sorted) list of words. This list could be compared against a standard glossary of, say, 50,000, words and represented by a 50,000-element binary vector with ones (1s) for the words that occurred in the document and zeros (0s) for those that did not. This vector could then be used to classify the document or in further analysis. This type of representation, which ignores the word order in the document, is sometimes called a bag of words representation.
to as a record or a tuple.
See Also: attribute, Machine Learning, Wise Wire.
Feedback
In general, this term is used to describe systems or inputs where the current output or state can modify the effect of input. A positive feedback acts as an amplifier or magnifier on the output (e.g., the rich get richer and the poor get poorer). A negative feedback acts to diminish large inputs and magnify small inputs. This
becomes important in keeping a system in control or "on target." Error-driven feedback systems use the deviation of the system from the current set point or
Page 107
target point to generate a corrective term. This concept is fundamental in the discussion of Robotics and control theory.
See Also: Robotics.
Feedback Network
A neural network is a feedback network if its graph contains any cycles.
Feedforward Network
A neural network is a feedforward network if the graph representing the network is acyclic (e.g., contains no cycles).
FFOIL
FFOIL is a specialization of the FOIL program specialized to learn functional relationships. In empirical trials on functions such as a greatest common denominator or an Ackermann's function, it has successfully been able to learn the function much faster than FOIL.
See Also: FOIL, Inductive Logic Programming.
Fifth Generation Computing
A term used by the Japanese to refer to their initiative to build a new generation of computers specially tuned for logic programming and logical inferences.
First-order learning is the process of learning a relationship from a database of positive and negatives
examples. It differs from the more common Machine Learning procedures, which learn attributes and values, in that it attempts to learn a generalizable relation. An example is the program FOIL.
See Also: FOIL, Inductive Logic Programming.
Fitness
In evolutionary and genetic algorithms, the fitness of a solution is a measure of how well the individual solution solves the task at hand. It can be used to select individual solutions for reproduction.
Page 108
Floyd's Shortest Distance Algorithm
This is one of several algorithms that can be used to find the shortest distance or lowest cost paths between nodes in graph. The connections and costs between the nodes are represented in an adjacency matrix. Floyd's algorithm constructs a cost matrix in n3 steps. Other algorithms can be used when this number is too large,
when only a few paths are needed, or when the costs are dynamic.
See Also: adjacency matrix, graph.
FOG
See: FOrecast Generator.
FOIL
FOIL is an inductive logic program that can learn first-order relationships. It uses a restricted form of Prolog, omitting cuts, fails, disjunctive goals and functions other than constants. It learns by using a divide-and-
conquer technique to expand clauses until no more examples can be absorbed, and can simplify definitions by pruning.
See Also: first-order learning, Logic Programming, Inductive Logic Programming, Prolog.
FOL
A program to check proofs stated in first order logic. Weyhrauch and Filman developed FOL at Stanford University in 1975.
F1 Layer
The initial resonating layer of an ART network.
FOrecast Generator (FOG) is a Canadian natural language generation system. It can translate weather forecasts from a database into either French or English.
See Also: generation, Natural Language Understanding.
Forest
A forest is a collection of trees. This term is sometimes used in discussion of techniques such as mixtures of experts, generalized addi-
Page 109
tive models, and boosting, that combine the results of a collection of decision trees to form a final decision.
See Also: boosting, generalized additive models, mixture-of-experts models.
FOR_RAIL
FOR_RAIL is a neural network-based crossing guard system under development by Nestor. It uses video sensors to collect input.
See Also: Artificial Neural Network.
FORTH
A low-level extensible stack-based programming language. It uses a reverse polish (or, postfix) syntax, so that the addition of two numbers would be described by the command sequence a b +, which would leave the resultant sum (a+b) on the top of the stack. Although basic FORTH is rather low-level, the language includes operations that allow the programmer to easily define new operations, as well as redefining existing
operations. Charles Moore designed FORTH for machine control in Astronomy, and has spread to numerous other areas, particularly in embedded systems. It has been used in mobile robots.
See Also: Mobile Robot.
Forward Chaining
A method of solving logic problems by working forward from the known data or previously proven inferences towards a goal or solution to a problem.
See Also: Backward Chaining.
Used in neural networks to mean prediction.
See Also: Artificial Neural Network.
Forward Reasoning
The process of reasoning from premises to conclusions. In automated logic systems, this can result in a rapid growth of conclusions that are irrelevant to the desired conclusion.
See Also: Backward Reasoning, logic programming, resolution.
Page 110
Fractal
A fractal is a compound object that contains multiple subobjects, each of which has some locally measurable characteristic similar to the same characteristic measured on the whole object. The ideas of fractals and fractal dimension are used in document and vision analysis.
See Also: Hausdorff dimension.
Frame of Discernment
The set of propositions that are of interest in Dempster-Shafer theory is referred to as the frame of discernment. This differs from the standard Universe that probability theory uses in that a frame of
discernment can include sets whose members are not in the frame of discernment. A frame of discernment can be refined by splitting apart its sets and can be coarsened by aggregating them. Two frames of discernment are compatible if they are equivalent after a process of refinement and/or coarsening.
Frame Representation Language (FRL)
Roberts and Goldstein of MIT developed the Frame Representation Language (FRL) in the late 1970s. The frame templates (classes) are organized into a hierarchy where the relationship between two objects is
described as ''a kind of". For example, persons would be a kind of mammal, which in turn might be a kind of mortal. Socrates would be an instance of the person class.
See Also: Frames.
objects in the knowledge base. The templates for an object can have default values. The slots can be further constrained by generic relationship (e.g., the age of a person is less than the age of the person's parents) and specific constraints for a particular object. The slots can also contain actions (functions) and goals. The frame for an object contains named "slots" for information about that object. The information in these slots can then be referred to in order to determine valid actions and goals. Minsky introduced frames as a form of organizing knowledge in 1975.
Page 111
See Also: Knowledge Base, Knowledge Representation, Semantic Memory.
Franz LISP
A LISP dialect implemented on the VAX machines. It was written in C and was, thus, portable to many UNIX machines.
Fredkin Prize
The Fredkin Prize was $100,000 prize for the first computer program to beat a reigning world chess champion. MIT Computer Science Professor Edward Fredkin established the prize in 1980. The inventors of the Deep Blue Chess machine won the Fredkin Prize. Deep Blue beat Gary Kasparov, the reigning world champ.
Professor Fredkin offered the final $100,000 as the third in a series of three prizes. Two scientists from Bell Laboratories (whose program first attained a masters rating in chess) won the first prize of $5,000. Five Carnegie Mellon graduate students who built Deep Thought (the first program to achieve international master status) claimed the second prize of $10,000.
See Also: Deep Blue.
FRA
See: Fuzzy Rule Approximation.
FRL
See: Frame Representation Language.
FSI
See: Fuzzy Singleton Inference; See Also: binary input-output fuzzy adaptive memory.
The second layer of an ART network where pattern choice and other behaviors take place.
See Also: ARTftp://ftp.sas.com/pub/neural/FAQ2.html, http://www.wi.leidenuniv.nl/art/.
Page 112
Function
A function is a relationship that takes zero or more attributes and returns a single item that may be compound. In class and frame based systems, a function that takes a single term and returns a single term is sometimes called a slot. A multivariate function is a function that returns a compound object, such as a list or an array, which may be regarded as the specification of a single multi-dimensional item.
See Also: anonymous, class, function, slot.
Functional Programming Languages
Functional programming languages are defined solely in terms of well-defined mathematical functions that take arguments and return values, with no side effects. In a pure functional language, there is no assignment; hence, the computation can be spread over many computers with reduced need for synchronization.
See Also: Logic Programming.
Function, Anonymous
A function that does not have a name, but is only defined inline with its use. Used in LISP as (lambda (args) (expression)), it allows for a more efficient code.
Function, Recursive
A function that can call itself while evaluating its arguments. A classic example is the factorial function f(x) := x f(x-1), for x>0 and f(x) = 1 for x < = 0. f(x) is the product of x and f(x-1), so that you must evaluate f(x-1), f(x-2), ... to evaluate f(x).
See Also: function, anonymous.
Fusion and Propagation
The fusion and propagation algorithm is a fundamental algorithm for producing multiple marginals from a graphical model that can be represented as a tree. It provides a rule for fusing incoming messages at each node and for propagating those messages out from a node. The fusion takes place in the local frame of discernment, so the full joint frame is never explicitly required.
Fuzzy ART
A network that synthesizes Adaptive Resonance Theory (ART) and Fuzzy Logic.
See Also:ftp:://ftp.sas.com/pub/neural/FAQ2.html, http://www.wi.leidenuniv.nl/art/.
Fuzzy ARTMAP
A supervised Fuzzy ART network.
See Also:ftp:://ftp.sas.com/pub/neural/FAQ2.html, http://www.wi.leidenuniv.nl/art/.
Fuzzy Associative Mmemory (FAM)
A fuzzy function, or model, that takes a k-dimensional fuzzy input and produces a 1-dimension fuzzy output. Comparable to a regression model or a neural network. The model can learn (estimate) from data or have its parameters set in some other fashion, e.g., by the model designer.
FuzzyCLIPS
FuzzyCLIPS is an enhanced version of C Language Integrated Production System (CLIPS) developed at the National Research Council of Canada that allows the implementation of fuzzy expert systems. It enables domain experts to express rules using their own fuzzy terms. It allows any mix of fuzzy and normal terms, numeric-comparison logic controls, and uncertainties in the rule and facts. Fuzzy sets and relations deal with fuzziness in approximate reasoning, while certainty factors for rules and facts manipulate the uncertainty. The use of the above modifications is optional and existing CLIPS programs still execute correctly.
See Also: C Language Integrated Production System, http://ai iit.nrc.ca/fuzzy/fuzzy.html.
Fuzzy Cognitive Map (FCM)
A graph with signed and directed arcs between nodes. The arc from node i to node j is an indicator of the influence of node i on node j. Likewise, the arc from j to i is an indicator of the influence of node j on node i. These arcs do not need to be symmetric. When the nodes are initialized to some state, repeated application of the connection matrix derived from the graph can be used to determine the evolution of the system and any fixed points or limit cycles.
Page 114
Fuzzy Cozmplement
The fuzzy complement of a fuzzy set is a fuzzy set whose membership functions are the complements of the original set (i.e., 1-m(A,x) where x are the elements in the original fuzzy set A).
Fuzzy Count
A measure of the size of a fuzzy set A and defined as the sum of all membership values in A. It is the fuzzy generalization of a classic set notion of cardinality, or size of a set.
Fuzzy Entropy
Fuzzy entropy is a measure of the fuzziness of a set. For a set A and its fuzzy complement B, fuzzy entropy is the ratio of the underlap of A B to the overlap of A B. In classical (crisp) sets, this value is, by
definition, 0, since the cardinality of A B is defined to be 0 and the cardinality of A B is similarly defined to be 1 (for a non-empty universe). The larger this value is, for a given set of elements, the fuzzier the set A (and B) becomes.
See Also: fuzzy set theory, underlap, overlap.
Fuzzy Intersection
In (crisp) set theory, the intersection of two sets is the set of all elements that are in both sets. In fuzzy set theory, the fuzzy intersection is the set of all elements with non-zero memberships in both sets. The
membership function of an element in this new set is defined to be the minimum of its membership in the two parent sets. Thus the intersection of a set and its complement (A ^ not-A), which is defined to be an empty set (of measure zero) in classic crisp sets, can be non-empty in fuzzy sets.
See Also: fuzzy complement, fuzzy set, fuzzy union.
Fuzzy Logic
A logic system based on manipulation of fuzzy sets. Some of the basic rules include definitions for intersections, unions, and complements.
Fuzzy Logic System
An inference system based on fuzzy logic.
Fuzzy Measure
The fuzzy measure of a fuzzy set is defined to be the sum of the fuzzy membership values of all elements with non-zero membership in the set. It is often denoted by m(A).
Fuzzy Rule Approximation (FRA)
A method for inference based on fuzzy logic. A given set of fuzzy rules defines a map from the input space to the output space. A FRA system attempts to replace the fuzzy rules with a neural network that approximates that rulebase.
Fuzzy Set
A set fuzzy set A consists of set objects X with membership values denoted by m(A,x). The relation that maps elements x in X to the membership values in (0,1) is called a membership function. A fuzzy set is a
generalization of classical set theory, where each element x in the universe X has a membership of 0 or 1 in a set A, and has been proposed as a means to deal with uncertainty.
See Also: fuzzy logic.
Fuzzy Singleton Inference (FSI)
See: Binary Input-Output Fuzzy Adaptive Memory.
Fuzzy Union
The fuzzy union of two sets is defined to be the set of all elements with non-zero memberships in either of the two sets. The membership function of elements in the new set is defined to be the maximum of the two
memberships in the parent sets.
See Also: fuzzy set.
FuzzyCLIPS
FuzzyCLIPS is an enhanced version of C Language Integrated Production System (CLIPS) developed at the National Research Council of Canada that allows the implementation of fuzzy expert systems. It enables domain experts to express rules using their own fuzzy terms. It allows any mix of fuzzy and normal terms, numeric-comparison logic controls, and uncertainties in the rule and facts. Fuzzy sets and relations deal with fuzziness in approximate reasoning, while certainty factors for rules and facts manipulate the uncer-
Page 116
tainty. The use of the above modifications is optional and existing CLIPS programs still execute correctly.
See Also: C Language Integrated Production System, http://ai.iit.nrc.ca/fuzzy/fuzzy.html.
Page 117
G
GA
See: Genetic Algorithm.
Gain
The figure of merit used to judge potential classification splits in ID3, defined as the change in entropy by the split. It was replaced with the gain ratio criterion in C4.5.
See Also: C4.5.
Gain Control
ART networks have two non-specific gain control signals, G1 and G2 for the F1 and F2 layers, respectively. G1 implements the 2/3 rule in the F1 layer, while G2 enables/disables the F2 layer.
See Also: ART, ftp:://ftp.sas.com/pub/neural/FAQ2.html, http://www.wi.leidenuniv.nl/art/.
Gain Ratio Criterion
A normalized version of the gain (entropy) criterion originally used in ID3. The latter tended to favor splits that generated many leaves. In C4.5, the gain score is divided by a split information score. This adjusts the gain from a split for the entropy of the number of splits.